Modified the language server so it sends the results of the interpreter

[emilkrebs]

This PR is needed to be merged and published on npm for the langium showcase:

eclipse-langium/langium-website#197

[msujew]

We should ensure that the interpreter doesn't run into an endless loop. For example, the following code will be quite problematic:

while (true) { }

The runInterpreter call should use a timeout (something like 5 seconds) and throw in case the execution didn't stop within the time limit.

[Lotes]

I think and suggest that you should avoid using UNKNOWN, because UNKNOWN cannot just be a primitive. PLUS, you can add more metadata if needed by trying this pattern:

type XType = 'A'|'B'|'C';
interface XBase {
  type: XType;
}

interface A extends XBase {
  type: 'A'; // subset of parent definition!
  content: string;
}

interface B extends XBase {
  type: 'B'; // subset of parent...
  content: number;
}

interface C extends XBase {
  type: 'C';
  content: boolean;
}
type X = A|B|C;

//usage: you can do stuff like this:
const x: X = ... //get from somewhere
if(x.type === 'A') {
   //it asserts then that x is of type A, so accessing x.content will be a string, no casts needed!!!
} else if(x.type === 'B') {
  //...x.content is number
} else if(x.type === 'C') {
  //...x.content is boolean
} else {
  assertUnreachable(x.type); //this code is then unreachable (it will be highlighted red when another type D was added)
}

[Lotes]

Maybe as separated PR, I see it is everywhere O.o...

[emilkrebs]

Hmm, maybe that would be better. What do you think @msujew?

[Lotes]

Minor: Coding style: It is normally a good habit to surround such pair calls like enter/leave with try-finally. Why? Because someone could accidentally write return within the section. Then (without try finally) the last part is not called. This can cause some debug time, which could be avoided by such a coding habit.
Are there any reasons against it?